آناتومی بحرانهای بانکی کشورهای در حال توسعه و بازار نوظهور

This paper uses a Binary Classification Tree (BCT) model to analyze banking crises in 50 emerging market and developing countries during 1990–2005. The BCT model identifies three conditions (and the specific threshold of the key indictors) at which the vulnerability to banking crisis increases—(i) very high inflation, (ii) highly dollarized bank deposits combined with nominal depreciation or low bank liquidity, and (iii) low bank profitability—which highlight that foreign currency risk, poor financial soundness, and macroeconomic instability are important drivers of banking crises. The results also emphasize the importance of conditional thresholds in triggering crises, in that banking crises are underlined by a combination of vulnerabilities—or a sequence of (non-linear) conditions—rather than the deterioration of a unique factor.

The advent of the 1990s witnessed a wave of banking crises in developing countries. These ranged from bank meltdowns in many transition economies in the early 1990s (triggered by macroeconomic instability), to the East Asian crises in 1997–1999 (spurred in part by over-lending), to the Dominican Republic crisis of 2003 (reflecting weak balance sheets). More recently, the 2008–2009 global financial crisis started in industrial countries but then spread to many developing and emerging market countries, including in Eastern Europe, straining financial systems that were characterized by excessive dollarized liabilities in the context of relative inflexible exchange rate regimes. Historically, banking crises have imposed a tremendous economic burden, including huge fiscal costs of resolution and/or sharp output losses.1 Consequently, the plethora of banking crises has sustained the drive for a better understanding of the factors that caused them. The extensive empirical literature on banking crises has generally used two standard econometric tools.2 The first is the signals approach, which studies and contrasts behaviors of economic indicators for periods both before and after a crisis, and identifies individual variables that best signal an impending crisis based on over- or under-shooting of specific threshold values (see Kaminsky and Reinhart, 1999). The second approach computes the probability of a banking crisis using a limited dependent variable model (see Demirgüc-Kunt and Detragiache, 1998 and Eichengreen et al., 1998). Both these tools consider the significance of individual factors in causing banking crises. In contrast, this paper provides a fresh perspective on banking crises, by demonstrating that such crises are most often underpinned by a combination of weaknesses rather than a single compelling factor. This paper makes an important contribution to the above literature by analyzing banking crises using a binary classification tree (BCT) technique, which to our knowledge has not been used previously to analyze banking crises.3 The BCT (described in detail below) is particularly useful to unravel the complex interactions between factors—for instance exchange rate depreciation (above a certain threshold), combined with large foreign exchange liabilities (above a certain threshold)—that eventually perpetuate a banking crisis. The BCT model also recognizes that economic indicators may have a non-linear impact on the probability of crisis, in that any increase or decrease of a key indicator need not increase crisis-proneness, unless the value of the indicator crosses a certain threshold, which the model identifies. The latter is very difficult to do in a standard regression analysis—while in principle one can always check the significance of variables at particular thresholds, it is virtually impossible to guess what these thresholds might be. Thus, using a BCT, we are able to identify first, the main indicators underlying a banking crisis; second, establish the threshold limits beyond which these indicators increase vulnerability to such a crisis; and third and most importantly, the combination of conditions between indicator variables underlying crisis-proneness. The focus of the literature, until now, has only been on addressing the first and in a few cases the second of the above questions. We feel that unraveling the complex pattern of relationships between indicator variables in the run up to a banking crisis is the most essential step to understanding such crises, which makes the BCT methodology a more appealing tool for analyzing banking crises compared to standard regression techniques. The BCT is a non-parametric statistical technique that is able to sift a large set of potential indicators and compare all candidate variables (at all possible threshold values) to identify which variables (and at what threshold values) are best able to split the sample and allocate the observations correctly into the two classes (in this case, crisis versus non-crisis). Thus, starting with the whole sample (parent node), two child nodes are generated at a particular threshold value of a key splitting variable such that the probability of crisis increases unambiguously in one child node and declines in the other, when compared to the probability of the crisis at the parent node. The process continues, in that each child node gets further split into more nodes, and eventually stops according to the criteria used to determine further improvements (see below). At each terminal node, the tree reveals a sequence of conditions among key indicators that can be identified as crisis-prone (and conversely also the conditions that describe a “tranquil” or non-crisis state). We analyze banking crises in a sample of 50 emerging market and developing countries during 1990–2005. The set of explanatory variables include: indicators of overall macroeconomic environment (growth, inflation, nominal depreciation, and government balance), external vulnerability (official foreign exchange (FX) reserve cover of broad money, export growth, and terms of trade (TOT) growth), monetary conditions (credit growth, real deposit rate, foreign interest rate, existence of explicit deposit insurance, and de facto exchange rate regime), and banking sector health (e.g., extent of liability dollarization in banks, net FX open position, bank liquidity, equity strength, asset quality, and two proxies for bank profitability). The baseline model identifies the following five candidate variables (out of the above 19) as most important determinants of banking crises: nominal depreciation, bank profitability, inflation, liability dollarization (given by the ratio of FX deposits to official FX reserves), and bank liquidity (given by the ratio of banks’ private credit to total deposits). It also identifies three key crises-prone conditions as described below, each of which reflects the combination of a “vulnerability” (or an underlying macroeconomic or banking sector weakness) with a “trigger” (or an adverse economic shock): • Macroeconomic instability: This crisis-prone condition is given by high annual inflation (greater than 19 percent) combined with a decline in the terms of trade (TOT) growth (less than 3¼ percent), which increases the probability of crisis from 5.3 percent—in the parent node—to 21.4 percent. High inflation could squeeze balance sheets by reducing real rates of return on assets (see Boyd and Champ, 2003), but banks become vulnerable under an excessively inflationary environment only (identified as inflation higher than 19 percent). Against this background, a drop in TOT growth below 3¼ percent—which affects the quality of banks’ trade credit and under normal circumstances could possibly be withstood easily, make banks almost four times more crisis prone (relative to the probability of crisis in the overall sample). The crises in Turkey during the 1990s and 2000 are identified as falling in this kind of banking crisis underpinned by macroeconomic instability. • Low profitability: This crisis-prone condition is given by low interest profitability (proxied by an interest rate spread of less than 3 percent) combined with a drop in export growth (less than 12 percent), which increases the probability of crisis to over 20 percent. Like condition (i), while a narrowing of interest margins is possibly not a bad thing per se (and could even be indicative of an increase in banking sector competition and financial deepening), it squeezes bank income after crossing a threshold (identified here as dropping below 3 percent), at which point, an export growth below 12 percent (a rather high threshold) results in a higher probability of crisis by deteriorating macroeconomic conditions more generally and the quality of bank assets exposed to export prospects more specifically. The banking crises in East Asia during the late 1990s (e.g., Indonesia, Malaysia, Korea, and Thailand during 1997–1998) are characterized by this kind of underlying vulnerability. • High foreign exchange (FX) risk: The final crisis-prone condition is given by high liability dollarization (FX deposits to official FX reserves more than 140 percent) combined with either, relatively high depreciation (greater than 9 percent), which raises the probability of crisis to 25 percent, or low bank liquidity (private credit to deposits higher than 150 percent), which raises the probability of crisis to 100 percent (although the latter condition is supported by only one observation). This confirms that FX risk—manifested as dollarization-induced liquidity risk—combined with a trigger such as high nominal depreciation increases proneness to banking crises. This supports the findings of Levy-Yeyati, 2005 and Levy-Yeyati, 2006 that financially-dollarized economies are generally more prone to banking crises, and that of Nicolo et al. (2003) who find that financial instability is likely higher in dollarized economies. Recent episodes of banking crises of this nature include Uruguay and the Dominican Republic in the early 2000s. Even more recently, the banking crises in many Eastern European countries in 2008–2009 (although outside the sample set of this paper) can also be characterized to be of this category. However, even when the exchange rate is stable, banks can still suffer from crises when their vulnerability from FX risk is combined with another weakness such as low bank liquidity, as demonstrated by the experience of Croatia in 1996. An alternative model that analyzes “severe” banking crises—defined as banking crises accompanied by negative real GDP growth—also identifies nominal depreciation (beyond 10 percent), liability dollarization (FX deposits beyond 179 percent of official FX reserves) and low bank liquidity (private credit to deposits more than 178 percent) as key precursors to banking crises. The basic BCT results also survive other robustness checks, confirming the importance of the BCT approach as a good practical tool for monitoring banking sector vulnerabilities. It is important to emphasize that while the BCT is a very useful technique to detect key indicators—and the non-linear relationship between them—that increases proneness to banking crises, these are definitely not the only possible explanations to all banking crises. Banking crises are complex and rare economic events, which have more often than not taken policymakers by surprise, reflecting the ample room that exists to further enhance existing early warning systems. This paper shows that the BCT makes a key contribution to the related literature by identifying various pre-conditions that have ignited recent crises in developing and emerging markets countries. The remaining paper is organized as follows. Section 2 briefly summarizes the existing literature on banking crises. Section 3 describes the BCT methodology and summarizes a relatively recent crisis literature that uses this approach. Section 4 presents the empirical analysis, while Section 5 concludes.

In this paper, we use a Binary Classification Tree (BCT) methodology to show that banking crises are underlined by a combination of “bads” and not any single compelling factor. The BCT model’s unique advantage is that it recognizes that an indicator may increase the probability of an outcome only after crossing a certain threshold, and sometimes only in combination with other conditions, and the model identifies both the thresholds and the sequence of relationships underlying the outcome of interest. This feature makes BCT a unique tool to analyze crises, which are generally triggered by a complex interaction of vulnerabilities. While the BCT has recently been used to analyze other economic crises including currency, sovereign debt and capital account crises, to our knowledge this is the first paper that uses the technique to look more closely at banking crises. To investigate the underlying vulnerabilities of a banking crisis we consider both traditional indicators of macroeconomic fundamentals and external factors, but also monetary conditions and financial soundness vulnerabilities. The BCT results stress that unconditional thresholds of variables are not as important in predicting banking crises as conditional thresholds, and that indicators trigger banking sector problems only after crossing these thresholds. Several indicators of financial market conditions are identified as important predictors of impending banking crises; in particular, nominal depreciation, interest profitability, liability dollarization and bank illiquidity. However, crises are triggered by complex interactions between financial soundness indicators and macroeconomic variables. The crisis-prone conditions identified by alternative BCT specifications confirm that foreign exchange risk is one of the largest vulnerabilities underlying banking crises, in that the probability of a banking sector crisis increases when preceded by a nominal depreciation combined with high liability dollarization. However, banks can be crisis-prone even when depreciation is limited, owing to the combination of high liability dollarization with low liquidity. For severe crises, defined as banking crises that are accompanied with negative growth, three indicators matter the most—nominal depreciation, liability dollarization in banks, and bank liquidity—but only when they cross identified thresholds. These results support the findings of recent literature that stress the role of FX risk in triggering banking crises (Kaminsky and Reinhart, 1999, Levy-Yeyati, 2005 and Levy-Yeyati, 2006, and Nicolo et al. (2003)). As a caveat, owing to the absence of time series data on these indicators, our analysis excluded some institutional indicators that have previously been associated with banking crises, such as lack of central bank independence, corporate governance weaknesses in the regulatory framework, and low share of foreign ownership of the banking system. When dummy variables corresponding to the thresholds of key primary splitters in the BCT are included in a standard logit regression they are statistically significant. In addition, dummy variables corresponding to the crisis terminal nodes of the BCT are also significant in the logit analysis. These findings confirm that banking crises are caused by a combination of macroeconomic, foreign exchange and financial soundness vulnerabilities acting together, which would have been virtually impossible to identify in a standard logit model. The ability of the BCT approach to identify these non-linear relationships underlying banking crises makes it a superior tool with which to monitor banking system vulnerabilities.